@empiricalrun/test-gen 0.58.0 → 0.59.0

package/CHANGELOG.md

@@ -1,5 +1,58 @@
 # @empiricalrun/test-gen
 
+## 0.59.0
+
+### Minor Changes
+
+- e2b0318: feat: dashboard session migration from CLI working
+- 894c5b5: fix: stricter types for tool call args in canonical message
+- 919e7df: fix: session create, duplicate user message, add Model Options
+- 045e370: feat: stateless build download for tool execution
+- f2d5859: feat: canonical migration for claude and openai
+- e2b0318: feat: migrate dashboard chat sessions to canonical chat state
+- 4d3b46a: feat: migrate to new chat state version in cli run
+
+### Patch Changes
+
+- 72f4577: feat: stop marking files as only and move to cmd builder approach
+- 82b3373: test: remove irrelevant or flaky tests
+- 3dbd4ff: fix: remove teardown skip since we interrupt agent with SIGINT
+- 7c77600: refactor: move test-gen file changes to prepare for removal
+- 562c25d: fix: update chat agent system prompt for using try/catch
+- 43656bf: refactor: split tool call execute into 2 methods
+- 51a9975: fix: unit tests for test-gen tool call
+- 772b93c: feat: gemini message transformer into canonical message type
+- 0c935ce: chore: chat model types are moved to shared-types pkg
+- 050fbd3: fix: model selection is ignored while creating a new session
+- ddfd009: fix: refactor chat agent and session list UI
+- 28682e2: feat: use test-run cmd builder in test-gen, remove test.only marking
+- 6b2ed6d: chore: use function args to tell if tool execution is remote
+- 511dbb1: test: skip tests for the old master agent
+- 7145e34: test: add new agent-loop test for openai canonical model
+- 43fb70f: feat: add retry functionality for chat sessions and update ChatAgent …
+- d1d1055: feat: integrate askUserForInput state into session handling
+- ef491c0: fix: incorrect build url getter
+- Updated dependencies [72f4577]
+- Updated dependencies [97cef16]
+- Updated dependencies [e2b0318]
+- Updated dependencies [772b93c]
+- Updated dependencies [894c5b5]
+- Updated dependencies [0c935ce]
+- Updated dependencies [919e7df]
+- Updated dependencies [045e370]
+- Updated dependencies [f2d5859]
+- Updated dependencies [919e7df]
+- Updated dependencies [ddfd009]
+- Updated dependencies [e2b0318]
+- Updated dependencies [28682e2]
+- Updated dependencies [b415e29]
+- Updated dependencies [09291a8]
+- Updated dependencies [aab69ae]
+- Updated dependencies [4d3b46a]
+- Updated dependencies [bc5996d]
+  - @empiricalrun/test-run@0.9.0
+  - @empiricalrun/llm@0.16.0
+
 ## 0.58.0
 
 ### Minor Changes

package/dist/agent/browsing/run.d.ts

@@ -1,4 +1,6 @@
 type GenerateTestsType = {
+    testCaseName: string;
+    testCaseSuites: string[];
     testFilePath: string;
     filePathToUpdate: string;
     pwProjectsFilter: string[];
@@ -7,7 +9,7 @@ type GenerateTestsType = {
     traceId?: string;
     editFileWithGeneratedCode: boolean;
 };
-export declare function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, pwProjectsFilter, testGenToken, repoDir, traceId, editFileWithGeneratedCode, }: GenerateTestsType): Promise<{
+export declare function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, testFilePath, filePathToUpdate, pwProjectsFilter, testGenToken, repoDir, traceId, editFileWithGeneratedCode, }: GenerateTestsType): Promise<{
     isError: boolean;
     error: string;
     actionsSummary?: string;

package/dist/agent/browsing/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAiBA,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC,CAuFD"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAcA,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC,CAsFD"}

package/dist/agent/browsing/run.js

@@ -4,13 +4,14 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.generateTestsUsingMasterAgent = generateTestsUsingMasterAgent;
+const test_run_1 = require("@empiricalrun/test-run");
 const detect_port_1 = __importDefault(require("detect-port"));
 const fs_1 = __importDefault(require("fs"));
 const web_1 = require("../../bin/utils/platform/web");
 const server_1 = require("../../file/server");
 const exec_1 = require("../../utils/exec");
 const utils_1 = require("./utils");
-async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, pwProjectsFilter, testGenToken, repoDir, traceId, editFileWithGeneratedCode, }) {
+async function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, testFilePath, filePathToUpdate, pwProjectsFilter, testGenToken, repoDir, traceId, editFileWithGeneratedCode, }) {
     if (!fs_1.default.existsSync(testFilePath)) {
         throw new Error(`File for master agent to run not found: ${testFilePath}`);
     }
@@ -32,24 +33,27 @@ async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, p
     const project = await (0, utils_1.detectProjectName)(testFilePath, playwrightConfig, pwProjectsFilter);
     const pageVar = await (0, web_1.getPageVariableNameFromCreateTest)(filePathToUpdate);
     console.log(`Detected playwright project name: ${project}`);
-    // run playwright test which will internally run the master agent
-    const teardownFileRegex = /.*\.teardown\.ts/;
-    const testsDirectory = `${repoDir}/tests`;
-    const isTestRunTriggeredForTeardown = teardownFileRegex.test(testFilePath);
-    const teardowns = new utils_1.TeardownManager(testsDirectory);
-    let removeListeners;
-    const command = `npx playwright test ${testFilePath} --retries 0 --project ${project} --timeout 0`;
-    let isError = false;
-    let error = "";
+    const testsDir = "tests";
+    const cmd = await (0, test_run_1.runSpecificTestsCmd)({
+        tests: [
+            {
+                name: testCaseName,
+                dir: testsDir,
+                filePath: testFilePath,
+                suites: testCaseSuites,
+            },
+        ],
+        projects: [project],
+        passthroughArgs: "--retries 0 --timeout 0",
+        //@ts-ignore
+        platform: "web",
+    });
+    let error = undefined;
     try {
-        if (!isTestRunTriggeredForTeardown) {
-            removeListeners = await teardowns.skipAll();
-        }
-        await pm.execute(command.split(" "), {
+        await pm.execute(cmd.command, cmd.args, {
             env: {
                 IPC_FILE_SERVICE_PORT: availablePort.toString(),
                 PW_TEST_HTML_REPORT_OPEN: "never",
-                // pass the test gen token so that the agent has the same configuration as cli
                 TEST_GEN_TOKEN: testGenToken,
                 PAGE_VAR_NAME: pageVar || "page",
                 DISPLAY: ":99",
@@ -60,15 +64,9 @@ async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, p
     catch (e) {
         error = `Failed to generate test using master agent ${e}`;
         console.error(error);
-        isError = true;
-    }
-    finally {
-        // Remove process listeners before unskipping files
-        removeListeners?.();
-        teardowns.unskipAll();
     }
-    // clean up the file if there is any error
-    if (isError) {
+    if (error) {
+        // clean up the file if there is any error
         try {
             const fileContent = fs_1.default.readFileSync(filePathToUpdate, "utf-8");
             const updatedContent = (0, web_1.replaceCreateTestWithNewCode)(filePathToUpdate, fileContent, "");
@@ -83,8 +81,8 @@ async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, p
     await (0, web_1.removeTestOnly)(testFilePath);
     await fileServer.stop();
     return {
-        isError,
-        error,
+        isError: !!error,
+        error: error || "",
         actionsSummary: fileServer.getActionsSummary(),
     };
 }

package/dist/agent/browsing/utils.d.ts

@@ -4,11 +4,7 @@ import { Page } from "playwright";
 import { PlaywrightTestConfig } from "playwright/test";
 export declare function isRegExp(obj: any): obj is RegExp;
 export declare function prepareBrowsingAgentTask(steps: string[]): string;
-export declare function replaceTodoWithCreateTest({ testFilePath, testCaseName, testCaseSuites, }: {
-    testFilePath: string;
-    testCaseName: string;
-    testCaseSuites: string[];
-}): Promise<void>;
+export declare function replaceTodoWithCreateTest(testFilePath: string): Promise<void>;
 export declare function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }: {
     testCaseName: string;
     testCaseSuites: string[];
@@ -33,13 +29,4 @@ export declare function getValidProjectNames(playwrightConfig: PlaywrightTestCon
  * @returns
  */
 export declare function detectProjectName(testFilePath: string, playwrightConfig: PlaywrightTestConfig, pwProjectsFilter?: string[]): Promise<string>;
-export declare class TeardownManager {
-    private directory;
-    constructor(directory: string);
-    private teardownFiles;
-    private getAllTeardownFiles;
-    private skipTeardownFile;
-    skipAll(): Promise<() => void>;
-    unskipAll(): void;
-}
 //# sourceMappingURL=utils.d.ts.map

package/dist/agent/browsing/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAIxE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAwBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,EAC9C,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B,iBA2BA;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,SAAS;gBAAT,SAAS,EAAE,MAAM;IACrC,OAAO,CAAC,aAAa,CAAqB;YAE5B,mBAAmB;YAUnB,gBAAgB;IAsBjB,OAAO;IAoBb,SAAS;CAKjB"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAI3D,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAsBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,YAAY,EAAE,MAAM,iBAsBnE;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB"}

package/dist/agent/browsing/utils.js

@@ -3,7 +3,6 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.TeardownManager = void 0;
 exports.isRegExp = isRegExp;
 exports.prepareBrowsingAgentTask = prepareBrowsingAgentTask;
 exports.replaceTodoWithCreateTest = replaceTodoWithCreateTest;
@@ -16,12 +15,10 @@ exports.detectProjectName = detectProjectName;
 const fs_1 = __importDefault(require("fs"));
 const minimatch_1 = require("minimatch");
 const path_1 = __importDefault(require("path"));
-const ts_morph_1 = require("ts-morph");
 // For TypeScript type safety
 let tsxImport = null;
 const logger_1 = require("../../bin/logger");
 const context_1 = require("../../bin/utils/context");
-const fs_2 = require("../../bin/utils/fs");
 const web_1 = require("../../bin/utils/platform/web");
 const create_test_block_1 = require("../codegen/create-test-block");
 const fix_ts_errors_1 = require("../codegen/fix-ts-errors");
@@ -96,7 +93,7 @@ async function prepareFileForUpdateScenario({ testCase, specPath, trace, }) {
     });
     return createTestFilePath;
 }
-async function replaceTodoWithCreateTest({ testFilePath, testCaseName, testCaseSuites, }) {
+async function replaceTodoWithCreateTest(testFilePath) {
     // This method is an alternative to prepareFileForUpdateScenario
     // TODO: Does not support scoped variables and updates in POM files
     const fileContent = fs_1.default.readFileSync(testFilePath, "utf-8");
@@ -109,11 +106,6 @@ async function replaceTodoWithCreateTest({ testFilePath, testCaseName, testCaseS
     const pageVariable = pageVarName || "page"; // Default to "page" if not specified
     fs_1.default.writeFileSync(testFilePath, fileContent.replace(todoRegex, (_, __, todoText) => `await createTest("${todoText.replace(/"/g, '\\"')}", ${pageVariable});`));
     await addImportForCreateTest(testFilePath);
-    await markTestAsOnly({
-        testCaseName,
-        testCaseSuites,
-        specPath: testFilePath,
-    });
 }
 async function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }) {
     const testFileContent = fs_1.default.readFileSync(specPath, "utf-8");
@@ -388,52 +380,3 @@ async function detectProjectName(testFilePath, playwrightConfig, pwProjectsFilte
     }
     return filteredProjectNames[0];
 }
-class TeardownManager {
-    directory;
-    constructor(directory) {
-        this.directory = directory;
-    }
-    teardownFiles = [];
-    async getAllTeardownFiles() {
-        const teardownFileRegex = /.*\.teardown\.ts/;
-        const teardownFiles = await (0, fs_2.readFilesInDirectory)(this.directory, (fileName) => teardownFileRegex.test(fileName));
-        return teardownFiles;
-    }
-    async skipTeardownFile(filePath) {
-        const project = new ts_morph_1.Project();
-        const sourceFile = project.addSourceFileAtPath(filePath);
-        sourceFile
-            .getDescendantsOfKind(ts_morph_1.SyntaxKind.CallExpression)
-            .forEach((callExpression) => {
-            const expression = callExpression.getExpression();
-            if (expression.getText() === "teardown") {
-                expression.replaceWithText("teardown.skip");
-            }
-            else if (expression.getText() === "test") {
-                expression.replaceWithText("test.skip");
-            }
-        });
-        // Save the modified file
-        sourceFile.save().catch((error) => {
-            console.error(`Error updating the file: ${filePath}`, error);
-        });
-    }
-    async skipAll() {
-        this.teardownFiles = await this.getAllTeardownFiles();
-        await Promise.all(this.teardownFiles.map(async ({ filePath }) => await this.skipTeardownFile(filePath)));
-        const setupProcessListeners = (cleanup) => {
-            const events = ["beforeExit", "exit", "SIGINT", "SIGTERM"];
-            events.forEach((event) => process.on(event, cleanup));
-            return () => {
-                events.forEach((event) => process.removeListener(event, cleanup));
-            };
-        };
-        return setupProcessListeners(this.unskipAll.bind(this));
-    }
-    unskipAll() {
-        this.teardownFiles.forEach(({ filePath, content }) => {
-            fs_1.default.writeFileSync(filePath, content, "utf-8");
-        });
-    }
-}
-exports.TeardownManager = TeardownManager;

package/dist/agent/chat/agent-loop.d.ts

@@ -3,12 +3,13 @@ import { IChatModel, SupportedChatModels } from "@empiricalrun/llm/chat";
 import { ToolCallService } from "../../tool-call-service";
 import { FileInfo } from "../../types";
 import { ReporterFunction } from "./types";
-export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, fileInfo, }: {
+export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, fileInfo, isToolExecutionRemote, }: {
     chatModel: IChatModel<any>;
     selectedModel: SupportedChatModels;
     toolCallService: ToolCallService;
     reporter: ReporterFunction;
     trace?: TraceClient;
     fileInfo: FileInfo;
+    isToolExecutionRemote: boolean;
 }): Promise<void>;
 //# sourceMappingURL=agent-loop.d.ts.map

package/dist/agent/chat/agent-loop.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,UAAU,EACV,mBAAmB,EAEpB,MAAM,wBAAwB,CAAC;AAGhC,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAGvC,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AAgB3C,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,EACL,eAAe,EACf,QAAQ,GACT,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,EAAE,eAAe,CAAC;IACjC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,QAAQ,EAAE,QAAQ,CAAC;CACpB,iBAoCA"}
+{"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAEL,UAAU,EACV,mBAAmB,EAEpB,MAAM,wBAAwB,CAAC;AAGhC,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAGvC,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AA0B3C,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,EACL,eAAe,EACf,QAAQ,EACR,qBAAqB,GACtB,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,EAAE,eAAe,CAAC;IACjC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,QAAQ,EAAE,QAAQ,CAAC;IACnB,qBAAqB,EAAE,OAAO,CAAC;CAChC,iBAuCA"}

package/dist/agent/chat/agent-loop.js

@@ -6,44 +6,61 @@ const picocolors_1 = require("picocolors");
 const prompt_1 = require("./prompt");
 const state_1 = require("./state");
 function getModelName(model) {
-    if (model.startsWith("claude"))
+    const provider = (0, chat_1.getProviderForModel)(model);
+    if (provider === "claude")
         return "Claude";
-    if (model.startsWith("gemini"))
+    if (provider === "google")
         return "Gemini";
-    if (model.startsWith("o4"))
+    if (provider === "openai")
         return "o4";
     return "AI";
 }
 const log = (...args) => {
     console.log((0, picocolors_1.gray)(args.join(" ")));
 };
-const isRemote = process.env.TOOL_EXECUTION_IS_REMOTE === "true" || false;
-async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, fileInfo, }) {
+function logError(context, error, trace) {
+    console.error((0, picocolors_1.gray)(`[Error in ${context}]:`), error instanceof Error ? error.stack || error.message : error);
+    trace?.update({
+        output: {
+            error: error instanceof Error ? error.message : String(error),
+        },
+    });
+}
+async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, fileInfo, isToolExecutionRemote, }) {
     const systemPrompt = await (0, prompt_1.buildSystemPrompt)(fileInfo);
     trace?.update({ input: { systemPrompt } });
     const { tools } = await toolCallService.getTools();
     while (!chatModel.askUserForInput) {
-        const toolCalls = chatModel.getPendingToolCalls();
-        if (toolCalls.length) {
-            const toolResults = await toolCallService.execute(toolCalls, isRemote, trace);
-            if (isRemote) {
-                log(`Tool call remote execution in progress`);
-                break;
+        try {
+            const toolCalls = chatModel.getPendingToolCalls();
+            if (toolCalls.length) {
+                if (isToolExecutionRemote) {
+                    await toolCallService.sendToQueue(toolCalls);
+                    log(`Tool call remote execution in progress`);
+                    break;
+                }
+                else {
+                    const toolResults = await toolCallService.execute(toolCalls, trace);
+                    chatModel.pushToolResultsMessage(toolCalls, toolResults);
+                }
+            }
+            log(`${getModelName(selectedModel)} is working...`);
+            const response = await chatModel.getLLMResponse({
+                systemPrompt,
+                tools: tools.map((tool) => (0, chat_1.zodToOpenAITool)(tool.schema)),
+                selectedModel,
+                trace,
+            });
+            if (!response) {
+                throw new Error("No response from LLM");
             }
-            chatModel.pushToolResultsMessage(toolCalls, toolResults);
+            chatModel.pushMessage(response);
+            const latest = chatModel.getHumanReadableLatestMessage();
+            await reporter((0, state_1.chatStateFromModel)(chatModel, selectedModel), latest);
         }
-        log(`${getModelName(selectedModel)} is working...`);
-        const response = await chatModel.getLLMResponse({
-            systemPrompt,
-            tools: tools.map((tool) => (0, chat_1.zodToOpenAITool)(tool.schema)),
-            selectedModel,
-            trace,
-        });
-        if (!response) {
-            throw new Error("No response from LLM");
+        catch (error) {
+            logError("chatAgentLoop", error, trace);
+            throw error;
         }
-        chatModel.pushMessage(response);
-        const latest = chatModel.getHumanReadableLatestMessage();
-        await reporter((0, state_1.chatStateFromModel)(chatModel, selectedModel), latest);
     }
 }

package/dist/agent/chat/exports.d.ts

@@ -1,10 +1,9 @@
-import { createChatModel, IChatModel, SupportedChatModels } from "@empiricalrun/llm/chat";
+import { createChatModel, IChatModel } from "@empiricalrun/llm/chat";
 import { FileInfo } from "../../types";
 import { chatAgentLoop } from "./agent-loop";
-import * as State from "./state";
-import { ChatStateOnDisk } from "./state";
+import { defaultModel, SUPPORTED_CHAT_MODELS, SupportedChatModels } from "./models";
+import { CHAT_STATE_VERSIONS_MIGRATIONS_MAP, chatStateFromModel, ChatStateOnDisk, createChatState, createChatStateForMessages, getLatestDownloadBuildUrl, LATEST_CHAT_STATE_VERSION, migrateChatState } from "./state";
 import { ReporterFunction } from "./types";
-declare const createChatState: typeof State.createChatState, createChatStateForMessages: typeof State.createChatStateForMessages, chatStateFromModel: typeof State.chatStateFromModel, loadChatState: typeof State.loadChatState, saveToDisk: typeof State.saveToDisk, CURRENT_CHAT_STATE_VERSION: string;
-export { chatAgentLoop, chatStateFromModel, createChatModel, createChatState, createChatStateForMessages, CURRENT_CHAT_STATE_VERSION, loadChatState, saveToDisk, };
+export { CHAT_STATE_VERSIONS_MIGRATIONS_MAP, chatAgentLoop, chatStateFromModel, createChatModel, createChatState, createChatStateForMessages, defaultModel, getLatestDownloadBuildUrl, LATEST_CHAT_STATE_VERSION, migrateChatState, SUPPORTED_CHAT_MODELS, };
 export type { ChatStateOnDisk, FileInfo, IChatModel, ReporterFunction, SupportedChatModels, };
 //# sourceMappingURL=exports.d.ts.map

package/dist/agent/chat/exports.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,UAAU,EACV,mBAAmB,EACpB,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,KAAK,KAAK,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AAE3C,QAAA,MACE,eAAe,gCACf,0BAA0B,2CAC1B,kBAAkB,mCAClB,aAAa,8BACb,UAAU,2BACV,0BAA0B,QACnB,CAAC;AAEV,OAAO,EACL,aAAa,EACb,kBAAkB,EAClB,eAAe,EACf,eAAe,EACf,0BAA0B,EAC1B,0BAA0B,EAC1B,aAAa,EACb,UAAU,GACX,CAAC;AAEF,YAAY,EACV,eAAe,EACf,QAAQ,EACR,UAAU,EACV,gBAAgB,EAChB,mBAAmB,GACpB,CAAC"}
+{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAErE,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,EACL,YAAY,EACZ,qBAAqB,EACrB,mBAAmB,EACpB,MAAM,UAAU,CAAC;AAClB,OAAO,EACL,kCAAkC,EAClC,kBAAkB,EAClB,eAAe,EACf,eAAe,EACf,0BAA0B,EAC1B,yBAAyB,EACzB,yBAAyB,EACzB,gBAAgB,EACjB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AAE3C,OAAO,EACL,kCAAkC,EAClC,aAAa,EACb,kBAAkB,EAClB,eAAe,EACf,eAAe,EACf,0BAA0B,EAC1B,YAAY,EACZ,yBAAyB,EACzB,yBAAyB,EACzB,gBAAgB,EAChB,qBAAqB,GACtB,CAAC;AAEF,YAAY,EACV,eAAe,EACf,QAAQ,EACR,UAAU,EACV,gBAAgB,EAChB,mBAAmB,GACpB,CAAC"}

package/dist/agent/chat/exports.js

@@ -1,48 +1,18 @@
 "use strict";
-var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    var desc = Object.getOwnPropertyDescriptor(m, k);
-    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
-      desc = { enumerable: true, get: function() { return m[k]; } };
-    }
-    Object.defineProperty(o, k2, desc);
-}) : (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    o[k2] = m[k];
-}));
-var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
-    Object.defineProperty(o, "default", { enumerable: true, value: v });
-}) : function(o, v) {
-    o["default"] = v;
-});
-var __importStar = (this && this.__importStar) || (function () {
-    var ownKeys = function(o) {
-        ownKeys = Object.getOwnPropertyNames || function (o) {
-            var ar = [];
-            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
-            return ar;
-        };
-        return ownKeys(o);
-    };
-    return function (mod) {
-        if (mod && mod.__esModule) return mod;
-        var result = {};
-        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
-        __setModuleDefault(result, mod);
-        return result;
-    };
-})();
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.saveToDisk = exports.loadChatState = exports.CURRENT_CHAT_STATE_VERSION = exports.createChatStateForMessages = exports.createChatState = exports.createChatModel = exports.chatStateFromModel = exports.chatAgentLoop = void 0;
+exports.SUPPORTED_CHAT_MODELS = exports.migrateChatState = exports.LATEST_CHAT_STATE_VERSION = exports.getLatestDownloadBuildUrl = exports.defaultModel = exports.createChatStateForMessages = exports.createChatState = exports.createChatModel = exports.chatStateFromModel = exports.chatAgentLoop = exports.CHAT_STATE_VERSIONS_MIGRATIONS_MAP = void 0;
 const chat_1 = require("@empiricalrun/llm/chat");
 Object.defineProperty(exports, "createChatModel", { enumerable: true, get: function () { return chat_1.createChatModel; } });
 const agent_loop_1 = require("./agent-loop");
 Object.defineProperty(exports, "chatAgentLoop", { enumerable: true, get: function () { return agent_loop_1.chatAgentLoop; } });
-const State = __importStar(require("./state"));
-const { createChatState, createChatStateForMessages, chatStateFromModel, loadChatState, saveToDisk, CURRENT_CHAT_STATE_VERSION, } = State;
-exports.createChatState = createChatState;
-exports.createChatStateForMessages = createChatStateForMessages;
-exports.chatStateFromModel = chatStateFromModel;
-exports.loadChatState = loadChatState;
-exports.saveToDisk = saveToDisk;
-exports.CURRENT_CHAT_STATE_VERSION = CURRENT_CHAT_STATE_VERSION;
+const models_1 = require("./models");
+Object.defineProperty(exports, "defaultModel", { enumerable: true, get: function () { return models_1.defaultModel; } });
+Object.defineProperty(exports, "SUPPORTED_CHAT_MODELS", { enumerable: true, get: function () { return models_1.SUPPORTED_CHAT_MODELS; } });
+const state_1 = require("./state");
+Object.defineProperty(exports, "CHAT_STATE_VERSIONS_MIGRATIONS_MAP", { enumerable: true, get: function () { return state_1.CHAT_STATE_VERSIONS_MIGRATIONS_MAP; } });
+Object.defineProperty(exports, "chatStateFromModel", { enumerable: true, get: function () { return state_1.chatStateFromModel; } });
+Object.defineProperty(exports, "createChatState", { enumerable: true, get: function () { return state_1.createChatState; } });
+Object.defineProperty(exports, "createChatStateForMessages", { enumerable: true, get: function () { return state_1.createChatStateForMessages; } });
+Object.defineProperty(exports, "getLatestDownloadBuildUrl", { enumerable: true, get: function () { return state_1.getLatestDownloadBuildUrl; } });
+Object.defineProperty(exports, "LATEST_CHAT_STATE_VERSION", { enumerable: true, get: function () { return state_1.LATEST_CHAT_STATE_VERSION; } });
+Object.defineProperty(exports, "migrateChatState", { enumerable: true, get: function () { return state_1.migrateChatState; } });

package/dist/agent/chat/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AACA,OAAO,EAGL,mBAAmB,EACpB,MAAM,wBAAwB,CAAC;AA0BhC,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAgGA;AA0BD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBAkDA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AACA,OAAO,EAGL,mBAAmB,EACpB,MAAM,wBAAwB,CAAC;AAmChC,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAiGA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBAsDA"}

package/dist/agent/chat/index.js

@@ -11,13 +11,14 @@ const file_tree_1 = require("../../utils/file-tree");
 const git_1 = require("../../utils/git");
 const agent_loop_1 = require("./agent-loop");
 const state_1 = require("./state");
+const DASHBOARD_DOMAIN = process.env.DASHBOARD_DOMAIN || "https://dash.empirical.run";
 function stopCriteria(userPrompt) {
     return userPrompt?.toLowerCase() === "stop";
 }
 function concludeAgent(chatModel, useDiskForChatState, selectedModel) {
     console.log(`\n${(0, picocolors_1.gray)("Usage summary -> " + chatModel.getUsageSummary())}`);
     if (useDiskForChatState) {
-        (0, state_1.saveToDisk)(chatModel.messages, selectedModel);
+        (0, state_1.saveToDisk)(chatModel.messages, selectedModel, chatModel.askUserForInput);
     }
 }
 async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialPromptContent, }) {
@@ -54,7 +55,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
     let userPrompt;
     let reporterFunc = async (chatState, latest) => {
         if (useDiskForChatState) {
-            (0, state_1.saveToDisk)(chatState.messages, selectedModel);
+            (0, state_1.saveToDisk)(chatState.messages, selectedModel, chatState.askUserForInput);
         }
         if (latest) {
             console.log(`${(0, picocolors_1.blue)(latest.role)}: ${latest.textMessage}`);
@@ -99,6 +100,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
                 trace,
                 toolCallService,
                 fileInfo,
+                isToolExecutionRemote: false,
             });
         }
     }
@@ -111,7 +113,6 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
     const usageSummary = chatModel.getUsageSummary();
     console.log(`\n${(0, picocolors_1.gray)("Usage summary -> " + usageSummary)}`);
 }
-const DASHBOARD_DOMAIN = process.env.DASHBOARD_DOMAIN || "https://dash.empirical.run";
 async function getChatSessionFromDashboard(chatSessionId) {
     const response = await fetch(`${DASHBOARD_DOMAIN}/api/chat-sessions/${chatSessionId}`, {
         headers: {
@@ -127,7 +128,11 @@ async function getChatSessionFromDashboard(chatSessionId) {
 }
 async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
     const chatSession = await getChatSessionFromDashboard(chatSessionId);
-    const chatState = chatSession.chat_state;
+    let chatState = chatSession.chat_state;
+    // If not already canonical, migrate to canonical format
+    if (!chatState.version || chatState.version !== state_1.LATEST_CHAT_STATE_VERSION) {
+        chatState = (0, state_1.migrateChatState)(chatState);
+    }
     const branchName = chatSession.branch_name;
     const trace = llm_1.langfuseInstance?.trace({
         id: chatSession.langfuse_trace_id,
@@ -164,5 +169,6 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
         trace,
         toolCallService,
         fileInfo,
+        isToolExecutionRemote: false,
     });
 }

package/dist/agent/chat/models.d.ts

@@ -0,0 +1,6 @@
+import type { ModelInfo } from "@empiricalrun/shared-types";
+export declare const SUPPORTED_CHAT_MODELS: readonly ModelInfo[];
+export type SupportedChatModels = (typeof SUPPORTED_CHAT_MODELS)[number]["id"];
+export declare const defaultModel: SupportedChatModels;
+export declare const modelLabels: Record<SupportedChatModels, string>;
+//# sourceMappingURL=models.d.ts.map

package/dist/agent/chat/models.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/models.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,4BAA4B,CAAC;AAE5D,eAAO,MAAM,qBAAqB,EAAE,SAAS,SAAS,EAqB5C,CAAC;AAEX,MAAM,MAAM,mBAAmB,GAAG,CAAC,OAAO,qBAAqB,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC;AAW/E,eAAO,MAAM,YAAY,EAAE,mBAA6C,CAAC;AAEzE,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,mBAAmB,EAAE,MAAM,CAOzD,CAAC"}

package/dist/agent/chat/models.js

@@ -0,0 +1,37 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.modelLabels = exports.defaultModel = exports.SUPPORTED_CHAT_MODELS = void 0;
+exports.SUPPORTED_CHAT_MODELS = [
+    {
+        id: "gemini-2.5-pro-preview-03-25",
+        label: "Gemini 2.5 Pro",
+        provider: "google",
+    },
+    {
+        id: "o4-mini-2025-04-16",
+        label: "OpenAI O4 Mini",
+        provider: "openai",
+    },
+    {
+        id: "claude-3-7-sonnet-20250219",
+        label: "Claude 3.7 Sonnet",
+        provider: "claude",
+    },
+    {
+        id: "claude-3-5-sonnet-20241022",
+        label: "Claude 3.5 Sonnet",
+        provider: "claude",
+    },
+];
+const DEFAULT_CHAT_MODEL_ID = "gemini-2.5-pro-preview-03-25";
+function getDefaultChatModelId() {
+    if (!exports.SUPPORTED_CHAT_MODELS.some((m) => m.id === DEFAULT_CHAT_MODEL_ID)) {
+        throw new Error("Default chat model is not in SUPPORTED_CHAT_MODELS");
+    }
+    return DEFAULT_CHAT_MODEL_ID;
+}
+exports.defaultModel = getDefaultChatModelId();
+exports.modelLabels = exports.SUPPORTED_CHAT_MODELS.reduce((acc, model) => ({
+    ...acc,
+    [model.id]: model.label,
+}), {});

package/dist/agent/chat/prompt.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/prompt.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAGvC,wBAAsB,iBAAiB,CAAC,QAAQ,EAAE,QAAQ,mBAoEzD"}
+{"version":3,"file":"prompt.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/prompt.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAGvC,wBAAsB,iBAAiB,CAAC,QAAQ,EAAE,QAAQ,mBAiGzD"}